2.1.

Subjects

Ten healthy subjects (age $32\pm 2.3$ years, five females) were recruited at the University of Zurich. Exclusion criteria were any psychiatric or neurological disorder or current medication. All subjects gave written informed consent. The study was approved by the ethics committee of the Canton Zurich (KEK-ZH-Nr: 2014-0056) and conducted in accordance with the Declaration of Helsinki.

2.2.

Experimental Protocol

Each subject underwent two experimental series separated by 1 week. Both experimental conditions were conducted at the same day and time between 9:00 and 12:00 am with a total duration of 10 min. Subjects were seated in a comfortable chair with the head positioned in a head rest in order to minimize head motion; this postural position was maintained during the whole experiment.

In both temperature conditions, subjects wore regular street wear and there was no other difference in the setup between the conditions. The following four respiratory challenges were assessed in each condition:

For BH and RB, subjects were trained prior to recording to perform the inspirational volume of air before the challenge similar to a normal breath cycle, in order to avoid inhaling a larger volume of air than the volume of a normal breath cycle.

2.3.

Functional Near-Infrared Spectroscopy Instrumentation

An NIRSport instrument (LLC NIRx Medical Technologies) was used for the fNIRS recordings. The system utilized time-multiplexed dual-wavelength light-emitting diodes. Each diode contained two light sources with wavelengths of 760 and 850 nm. Optical detection was performed with photoelectrical detectors containing silicon photodiodes (Siemens, Germany). Sources and detectors were placed in a head cap to allow for direct skin contact (Epitex Inc., Japan). The data acquisition board was connected to a notebook computer running LabVIEW 2011 (National Instruments, Austin, Texas). The fNIRS data were recorded with a sampling frequency of 7.81 Hz. The probe setup covered parts of the prefrontal cortex (Fig. 1). Functional recordings were visually inspected for motion artifacts (in particular, “steps” and “spikes”) without the need for removal of artifacts. The time series of oxyhemoglobin (${\mathrm{O}}_2\mathrm{Hb}$) were then used to extract HRV (Sec. 2.4).

Fig. 1

fNIRS channel positions. Channel setup covering the prefrontal cortex (channels indicate the middle between source and detector). The MATLAB^® toolbox NFRI²⁸ was used to estimate the MNI coordinates of the used EEG 10 to 20 position. Channel positions were visualized using BrainNet Viewer.²⁹

2.4.

Pulse Rate Variability Extraction from Functional Near-Infrared Spectroscopy

${\mathrm{PRV}}_{\mathrm{NIRS}}$ was extracted using the automatic multiscale-based peak detection (AMPD) algorithm developed by Scholkmann et al.³⁰ AMPD is based on the calculation and analysis of the local HR maxima in the raw fNIRS time series. AMPD detects the HR peaks, which are then used to calculate the interpeak intervals frequency via interpolating the time difference signal. The first step of the AMPD algorithm consists of calculating the local maxima scalogram (LMS). To this end, the signal is first linearly detrended in that the least-squares fit of a straight line to a given raw ${\mathrm{O}}_2\mathrm{Hb}$ signal is calculated and subtracted from the signal. The local maxima of the signal are then determined using a moving window approach. The second step of the algorithm comprises a row-wise summation of the LMS matrix resulting in a vector $v$. The global minimum $\lambda$ of the vector $v$ represents the scale with the most local maxima, which is then used in the third step to reshape the LMS matrix by removing all elements larger than $\lambda$. In the last step of the algorithm, the HR peaks are detected by calculating the columnwise standard deviation of the LMS matrix. Each point of the total number of detected peaks of a given signal indicated the values of the detected peaks (Fig. 2). The AMPD calculation was done for each of the channels (i.e., 1 to 16) and for each subject. Channels without cardiac components due to noise were identified and excluded from analysis. A final ${\mathrm{PRV}}_{\mathrm{NIRS}}$ estimate was obtained for each single subject by computing the median PRV over all channels at each time point.

Fig. 2

PRV-extraction from fNIRS. Illustration of the results of applying the AMPD algorithm³⁰ to raw fNIRS time series to calculate ${\mathrm{PRV}}_{\mathrm{NIRS}}$. Red dots represent the detected HR peaks in a given signal with the peaks indicating the values of PRV at each point.

2.5.

Photoplethysmography-Derived Pulse Rate Variability

A LifeSense LS1-9R multiparameter instrument (Nonin Medical, Sweden) was used to derive HR from finger PPG. Subjects wore a finger pulse oximeter through which HR were sampled at 10 Hz. Triggers were logged to allow for temporal synchronization with the fNIRS data. For statistical analysis, HR values were automatically computed by the LifeSense instrument from the raw data.

3.1.

Accuracy Between Pulse Rate Variability Measures

3.2.

Sympathetic and Parasympathetic Autonomous Nervous System Contributions on Pulse Rate Variability

The following measures of short-term and long-term variabilities were assessed only in the RS series. The respiratory challenges were analyzed only with respect to the short-term variability (see rationale in Secs. 3.2.1 and 3.2.2).

4.1.

Pulse Rate Variability Near-Infrared Spectroscopy and Photoplethysmography

Extraction of ${\mathrm{PRV}}_{\mathrm{NIRS}}$ was performed for each of the fNIRS channels 1 to 16 per subject. 13% of the channels over all subjects did not contain cardiac components due to a low signal-to-noise ratio and were excluded from analysis. A single-subject example of the final ${\mathrm{PRV}}_{\mathrm{NIRS}}$ estimate obtained by computing the median PRV over all channels at each time point is shown in Fig. 3(a) in comparison to ${\mathrm{PRV}}_{\mathrm{PPG}}$.

Fig. 3

${\mathrm{PR}}_{\mathrm{NIRS}}$ and ${\mathrm{PR}}_{\mathrm{PEG}}$. (a) Single-subject examples illustrating raw time courses of ${\mathrm{HR}}_{\mathrm{NIRS}}$ and ${\mathrm{HR}}_{\mathrm{PPG}}$. (b) Group-level beta estimates of HR responses to respiratory challenges, HV, BH, and RB. Error bars indicate standard error of the mean (SEM). Significance between temperature conditions was highlighted (*).

Using a general linear model, the beta estimates of the HR in response to the respiratory challenges were calculated per subject and shown on the group-level [Fig. 3(b)]. Results showed that compared to RS, HV (${\mathrm{PRV}}_{\mathrm{NIRS}}$: $p<0.001$, ${\mathrm{PRV}}_{\mathrm{PPG}}$: $p<0.046$) and RB (${\mathrm{PRV}}_{\mathrm{NIRS}}$: $p<0.009$, ${\mathrm{PRV}}_{\mathrm{PPG}}$: $p<0.0001$) induced a larger increase in HR in the CONTROL condition compared to the COLD condition, whereas BH (${\mathrm{PRV}}_{\mathrm{NIRS}}$: $p=0.044$, ${\mathrm{PRV}}_{\mathrm{PPG}}$: $p<0.0001$) induced a smaller HR decrease in condition CONTROL compared to condition COLD. These results indicated that the HR response was overall diminished in the COLD compared to the CONTROL condition.

4.2.

Accuracy Between Pulse Rate Variability Measures

The overall correlation between ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ was $r>0.9$. The accuracy between ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ was assessed using C-COV and C-SampEn (Fig. 4). Repeated measures ANOVA showed a consistent pattern with significant main effects for both factors “temperature” and “respiration” (Table 1). The between-subject factor “temperature” (CONTROL versus COLD) revealed a smaller main effect for C-COV ($F=10.026$, $p=0.005$) compared to C-SampEn ($F=52.134$, $p<0.0001$). The within-subject factor “respiration” (RS, HV, BH, and RB) revealed larger C-COV correlation coefficients between ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ in the RS in the CONTROL ($r=0.903$) compared to the COLD ($r=0.803$) condition. During the respiratory challenges, correlation coefficients decreased with the smallest values in response to BH under both the CONTROL and the COLD condition. Sample entropy represented a mirrored pattern of the correlation analysis, but with an overall larger “temperature” effect, but smaller effects of “respiration” indicating less variance between the respiratory challenges.

Fig. 4

C-COV and C-SampEn. Group-level results of the accuracy between ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ assessed using C-COV and C-SampEn. Error bars indicate SEM. Significant differences between temperature conditions were highlighted (*). See Table 1 for ANOVA.

Table 1

ANOVA C-COV and C-SampEn. Repeated measures ANOVA assessing the accuracy between PRVNIRS and PRVPPG using C-COV and C-SampEn, with the between-subject factor “temperature” (CONTROL versus COLD) and the within-subject factor “respiration” (RS, HV, BH, and RB), followed by post-hoc comparisons using paired-test. See Fig. 4 for illustration.

4.3.

Sympathetic and Parasympathetic Autonomous Nervous System Contributions on Pulse Rate Variability

4.3.1.

Frequency-domain (low frequency/high frequency ratio)

To investigate the frequency aspects that may characterize the contributions of the SNS and PSNS, the HF (0.15 to 0.4 Hz) and LF (0.04 to 0.15 Hz) activities were assessed. Results from the RS indicated that the COLD condition elicited an overall smaller LF/HF ratio compared to the CONTROL condition (Fig. 5). The difference was significant only for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ ($p<0.018$), but not for ${\mathrm{PRV}}_{\mathrm{PPG}}$ ($p=0.829$). The proportional change from the CONTROL to COLD condition was larger in HF activity (76.48%) compared to LF activity (25.48%) (Table 2), indicating that higher HF activity under the COLD condition was the main proportional contributor to the statistical difference. Results obtained from the respiratory challenges showed that the difference in HF activity reached statistical significance only for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ for all challenges, but not for ${\mathrm{PRV}}_{\mathrm{PPG}}$ (Fig. 6).

Fig. 5

${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ measures (RS). (a) Single-subject level. Examples of the PRV measures assessed in the frequency-domain, the time-domain, the Poincaré plot, and the CCM. (b) Group-level. Bar graphs illustrating the $\text{mean}\pm \mathrm{SEM}$ of the LF/HF ratio, the SDNN/RMSSD ratio, the SD1/SD2 ratio, and the CCM. Significant differences between temperature conditions were highlighted (*). See Table 2 for statistics.

Table 2

PRVNIRS and PRVPPG measures (RS). Mean/median±SEM of LF, HF, SDNN, RMSSD, SD1, SD2, and CCM. Differences between the CONTROL and the COLD condition were assessed using paired t-test. Units are given in milliseconds.

			CONTROL			COLD			T-test		Proportional change (mean)
			Mean	SEM	Median	Mean	SEM	Median	F	p-value
Frequency-domain	${\mathrm{PRV}}_{\mathrm{NIRS}}$	LF	2852.753	588.569	2514.700	3579.560	640.651	3093.000	0.698	0.414	25.48%
		HF	156.343	34.445	130.115	275.921	58.104	217.925	3.134	0.094	76.48%
	${\mathrm{PRV}}_{\mathrm{PPG}}$	LF	484.164	80.638	390.130	477.207	137.414	328.095	0.002	0.966	$-1.44\%$
		HF	21.344	1.531	21.422	22.738	6.747	12.380	0.041	0.843	6.53%
	${\mathrm{PRV}}_{\mathrm{NIRS}}$	LF/HF ratio	18.399	1.090	1869.181	14.016	1.275	1348.546	6.826	0.018
	${\mathrm{PRV}}_{\mathrm{PPG}}$	LF/HF ratio	22.085	2.903	1851.050	21.254	2.426	2331.723	0.048	0.829
Time-domain	${\mathrm{PRV}}_{\mathrm{NIRS}}$	SDNN	102.090	13.579	82.350	106.420	16.210	106.250	0.042	0.840	4.24%
		RMSSD	3.330	0.339	3.300	4.250	0.508	3.650	2.271	0.149	27.63%
	${\mathrm{PRV}}_{\mathrm{PPG}}$	SDNN	40.080	5.714	37.150	40.540	3.556	37.950	0.005	0.946	1.15%
		RMSSD	1.180	0.106	1.133	1.260	0.138	1.150	0.212	0.651	6.78%
	${\mathrm{PRV}}_{\mathrm{NIRS}}$	SDNN/RMSSD ratio	30.218	1.772	28.843	24.533	1.884	22.755	4.834	0.041
	${\mathrm{PRV}}_{\mathrm{PPG}}$	SDNN/RMSSD ratio	33.117	1.886	31.217	34.106	3.386	32.025	0.065	0.801
Poincaré plot	${\mathrm{PRV}}_{\mathrm{NIRS}}$	SD1	0.003	0.000	0.003	0.004	0.001	0.003	0.571	0.460	18.22%
		SD2	0.133	0.011	0.119	0.114	0.020	0.079	0.705	0.412	$-14.31\%$
	${\mathrm{PRV}}_{\mathrm{PPG}}$	SD1	0.001	0.000	0.001	0.001	0.000	0.001	0.007	0.932	$-1.52\%$
		SD2	0.052	0.006	0.048	0.055	0.004	0.054	0.179	0.678	5.87%
	${\mathrm{PRV}}_{\mathrm{NIRS}}$	SD1/SD2 ratio	0.022	0.002	0.023	0.032	0.003	0.031	11.204	0.004
	${\mathrm{PRV}}_{\mathrm{PPG}}$	SD1/SD2 ratio	0.018	0.001	0.017	0.017	0.002	0.015	0.164	0.690
CCM	${\mathrm{PRV}}_{\mathrm{NIRS}}$	CCM	0.003	0.000	0.003	0.005	0.001	0.005	4.492	0.047
	${\mathrm{PRV}}_{\mathrm{PPG}}$	CCM	0.004	0.000	0.004	0.004	0.000	0.004	0.062	0.806

Fig. 6

${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ measures (respiratory challenges). Bar graphs illustrating the $\text{mean}\pm \mathrm{SEM}$ of the short-term variability measures, HF, RMSSD, and SD1, during HV, BH, and RB, for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$. Significant differences between temperature conditions were highlighted (*).

4.3.5.

Receiver operating characteristic for pulse rate variability measures

To illustrate the performance of ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ in predicting the temperature change, ROC curves were generated based on logistic regression with the binary classifier “temperature” (CONTROL versus COLD) and the PRV measures as response variables. The AUC values obtained from the RS showed that ${\mathrm{PRV}}_{\mathrm{NIRS}}$ predicted the temperature change better than ${\mathrm{PRV}}_{\mathrm{PPG}}$ for all measures, i.e., the LF/HF ratio, the SDNN/RMSSD ratio, the SD1/SD2 ratio, and the CCM (Fig. 7). The differences between AUC reached significance level only for the SD1/SD2 ratio $p<0.0001$ (LF/HF ratio $p=0.054$, SDNN/RMSSD ratio $p=0.069$, and CCM $p=0.224$). Results obtained from the respiratory challenges confirmed these findings, showing that the AUC values for HF activity, RMSSD, and SD1 were larger for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ for most of the challenges compared to ${\mathrm{PRV}}_{\mathrm{PPG}}$ (exceptions see Fig. 8).

Fig. 7

ROC for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ (RS). Group-level ROC curves and AUC values based on logistic regression with the binary classifier “temperature” (CONTROL versus COLD) for each of the PRV measures, the LF/HF ratio, the SDNN/RMSSD ratio, the SD1/SD2 ratio, and the CCM. Statistical significance of AUC-differences: LF/HF-ratio $p=0.054$, SDNN/RMSSD ratio $p=0.069$, SD1/SD2 ratio $p<0.0001$, CCM $p=0.224$.

Fig. 8

ROC for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ (respiratory challenges). Group-level ROC curves and AUC values based on logistic regression with the binary classifier “temperature” (CONTROL versus COLD) for each of the short-term variability measures, HF, RMSSD, SD1, during HV, BH, and RB, for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$. Statistical significance of AUC-differences in HV: HF $p=0.009$, RMSSD $p=0.015$, SD1 $p=0.011$; statistical significance of AUC-differences in BH: HF $p<0.0001$, RMSSD $p<0.0001$, SD1 $p<0.0001$; statistical significance of AUC-differences in RB: HF $p=0.285$, RMSSD $p=0.091$, SD1 $p<0.0001$.

4.3.6.

Correlation between pulse rate variability measures

Pearson correlation coefficients were computed between LF, HF, SDNN, RMSSD, SD1, SD2, and CCM values, for the COLD and CONTROL condition (Fig. 9). Results confirmed that the indices of short-term variability (HF, RMSSD, and SD1) and long-term variability (LF, SDNN, and SD2) exhibited a strong positive correlation each other, and a negative correlation with CCM. The correlations were less stable for ${\mathrm{PRV}}_{\mathrm{PPG}}$ under the COLD condition. Results obtained from the respiratory challenges are shown in the Fig. 10.

Fig. 9

Correlation between PRV measures (RS). Pearson correlation coefficients for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ between LF, HF, SDNN, RMSSD, SD1, SD2, and CCM values, for the CONTROL and the COLD conditions.

Fig. 10

Correlation between PRV measures (respiratory challenges). Pearson correlation coefficients for ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ between the short-term variability measures, HF, RMSSD, and SD1, for the CONTROL and the COLD conditions.

5.1.

Accuracy Between Pulse Rate Variability Measures

To compare the accuracy between ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$, the present analysis assessed C-COV and C-SampEn. C-COV showed the relative correlations between the ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$, whereas C-SampEn reflected the regularity between the two PRV measures. Overall, we found good agreement between the two methods with the largest coefficients under RS ($r=0.903$). However, both changes in respiration and temperature resulted in reduced correlation (C-COV) and stronger irregularity (C-SampEn) between ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ (Fig. 4 and Table 1). The lower correlation in response to respiratory and temperature changes may reflect the previously reported inaccuracy of ${\mathrm{PRV}}_{\mathrm{PPG}}$ under nonoptimal conditions, such as including motion artifacts,⁶ physical strenuous exercise,⁵ or mental stressors.⁷ Although our subjects were instructed to keep the body as still as possible during the respiratory challenges, the breathing and temperature changes certainly induced minor body motions and mental stress, respectively, that may have contributed to the inaccuracy of ${\mathrm{PRV}}_{\mathrm{PPG}}$. The ${\mathrm{PRV}}_{\mathrm{NIRS}}$ signal shows more high-frequency variability compared to ${\mathrm{PRV}}_{\mathrm{PPG}}$. That this high-frequency variability contains some physiological relevant information (i.e., diminished in the ${\mathrm{PRV}}_{\mathrm{PPG}}$ data) is a shown by the present results.

It should further be mentioned that the correlations between ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ were slightly lower than that has been reported previously in adult²⁵ and infant data.²⁶ These differences may arise due the difference between PPG (applied in our study) versus EEG (applied in the two other studies).

5.2.

Sympathetic and Parasympathetic Autonomous Nervous System Contributions on Pulse Rate Variability

To characterize the contributions of the SNS or PSNS on ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$ in response to the temperature change, we applied frequency- and time-domain measures based on linear and nonlinear methods.

In general, exposure to cold temperature changes the relationship between the SNS or PSNS. Whereas the PSNS (the “rest and digest” system) restores the body’s energy primarily associated with decreases in PRV, the SNS (the “fight and flight” system) is primarily associated with increases in PRV. Furthermore, although the effect on PRV of cooling varies with the duration of the temperature change, it is thought that cold exposure modulates short-term variability (i.e., HF, RMSSD, and SD1) more than long-term variability (LF, SDNN, and SD2).⁸ This indicates that cold exposure modulates SNS-PSNS balance by causing a shift toward increased PSNS activity. Previous studies comparing PRV derived from PPG and ECG under optimal rest conditions did not report significant differences in PRV assessment.^6,45,46 However, studies that compared PPG and ECG under less optimal conditions have shown that ${\mathrm{PRV}}_{\mathrm{PPG}}$ assessed the parameters of short-term variability (HF, RMSSD, and SD1) considerable less reliable compared to ECG.^7,9

In line with these findings, our results showed that the cold temperature change (COLD) exhibited significant increases of the parameters reflecting short-term variability (i.e., HF activity, RMSSD, and SD1), whereas the parameters reflecting long-term variability did not change significantly compared to the control condition (CONTROL). Importantly, these differences between the cold and control temperature could only be significantly characterized by ${\mathrm{PRV}}_{\mathrm{NIRS}}$, but not by ${\mathrm{PRV}}_{\mathrm{PPG}}$. Furthermore, the better performance in characterizing short-term variability of ${\mathrm{PRV}}_{\mathrm{NIRS}}$ compared to ${\mathrm{PRV}}_{\mathrm{PPG}}$ was not only observed under potentially stressful respiratory challenges (Figs. 6 and 8), but were even detectable under RS conditions (Fig. 5 and Table 2). The ROC analysis confirmed these results indicating that ${\mathrm{PRV}}_{\mathrm{NIRS}}$ had a better predictive power to differentiate the temperature change compared to ${\mathrm{PRV}}_{\mathrm{PPG}}$, with the SD1/SD2 ratio obtained from the Poincaré plot revealing the best discrimination (AUC-/ROC-differences, Fig. 7).

Together, our results indicated that ${\mathrm{PRV}}_{\mathrm{NIRS}}$ had a significantly higher sensitivity to the short-term contributions of the PSNS compared to ${\mathrm{PRV}}_{\mathrm{PPG}}$. These findings show that under conditions which may induce changes primarily in short-term variability, both research and clinical settings may benefit from using ${\mathrm{PRV}}_{\mathrm{NIRS}}$ to overcome the lower reliability of ${\mathrm{PRV}}_{\mathrm{PPG}}$. That the ${\mathrm{PRV}}_{\mathrm{NIRS}}$ signal contains more high-frequency information than ${\mathrm{PRV}}_{\mathrm{PPG}}$ can also be easily see in Fig. 3.

5.3.

Methodological Considerations and Limitations

Core and skin temperature are the body’s two temperature components. Core temperature ($T_{\mathrm{c}}$) represents the internal or deep body temperature, whereas skin temperature represents the mean outside surface body temperature ($T_{\mathrm{sk}}$). The average temperature of the body at any time is a weighted balance between these two temperature components. When confronted with thermal stress (heat or cold), the body strives to control $T_{\mathrm{c}}$ through physiological adjustments, and $T_{\mathrm{sk}}$ provides the major feedback to the brain to coordinate this control.

While $T_{\mathrm{sk}}$ varies greatly with ambient temperature, $T_{\mathrm{b}}$ is relatively stable. When $T_{\mathrm{sk}}$ decreases, HR in general goes down due to a parasympathetic reflex.⁴⁷ However, if the cold stress is of sufficient magnitude to decrease $T_{\mathrm{c}}$, HR may either increase (due to sympathetic activation) or decrease (due to increased central blood volume). Regarding the present study, we hypothesized that the cold temperature change was not strong enough to affect $T_{\mathrm{c}}$ (although, this hypothesis could not be confirmed, since we did not directly assess $T_{\mathrm{sk}}$ or $T_{\mathrm{c}}$). Therefore, the induced cold temperature change in this study was hypothesized to decrease $T_{\mathrm{sk}}$ only. In line with this assumption, we found that HR amplitudes were significantly reduced in response to the cold temperature change (Fig. 3), most likely reflecting a general HR slowdown due to the assumed increase in PSNS activity.^8,47

5.4.

Conclusion

By comparing simultaneously acquired ${\mathrm{PRV}}_{\mathrm{NIRS}}$ and ${\mathrm{PRV}}_{\mathrm{PPG}}$, the present study showed that ${\mathrm{PRV}}_{\mathrm{NIRS}}$ may outperform ${\mathrm{PRV}}_{\mathrm{PPG}}$ under conditions involving respiratory and temperature changes. In particular, our results indicated that ${\mathrm{PRV}}_{\mathrm{NIRS}}$ may provide a better characterization of the short-term contributions of the autonomous nervous system compared to ${\mathrm{PRV}}_{\mathrm{PPG}}$. Extracting ${\mathrm{PRV}}_{\mathrm{NIRS}}$ may, therefore, be advantageous for both research and clinical settings, while being a brain imaging method at the same time.